A strong Physical AI company needs some combination of: 1. Painkiller use case: expensive, dangerous, repetitive, dirty or labor-constrained work. 2. Clear ROI: customer payback measurable in months, not an abstract productivity story. 3. High utilization: a $100k robot working 2 hours/day has very different economics from one working 18. 4. Reliability: demo success ≠ production reliability. 5. Low intervention rate: constant teleoperation can destroy the economics. 6. Deployment velocity: how quickly can you go from signed customer → productive robot? 7. Uptime / MTBF: how often does something physically break? 8. Field service: who repairs it, how quickly, and at what cost? 9. Deployment repeatability: does every new customer require a custom engineering project? 10. Workflow integration: how much does the customer need to change around the robot? 11. Safety: what is the cost of failure in the real world? 12. Data flywheel: does every deployment make the system better? 13. Fleet learning: can what robot #100 learns improve robot #1,001? 14. Gross margin after reality: maintenance, installation, teleoperation, spare parts and on-site support all count. 15. Form-factor advantage: why does this task need this embodiment at all? The biggest Physical AI companies won’t be built on the best demos. They’ll be built on the best real-world economics.